Allow preferred docs reviewers

[ptgott]

The code reviewer configuration allows us to name preferred reviewers for certain file paths. For docs PRs, the workflow bot does not consult the preferred reviewer configuration. This means that, even if we add the preferredReviewerFor key to a codeReviewers entry with a docs path as a value, the workflow bot continues to add only docsReviewers to docs PRs.

In this change, the workflow bot gets docs reviewers by:

1. Retrieving the docs reviewer pool from the config (like it does now).
2. Retrieving all preferred reviewers for the file paths included within a PR. Since the pool of dedicated docs reviewers is small, the workflow bot gets all preferred reviewers for a file path, not just one.
3. Merging the two lists of reviewers.

The alternative would be to allow preferred file paths in the docs reviewer pool, but it makes sense to have a small, consistent set of docs reviewers who have experience with docs reviews and a larger pool of subject matter experts for specific file paths from the code reviewer pool.

Implementation notes:

• Adds an *Assignments.getAllPreferredReviewers method, which fetches all preferred reviewers for all files in a change set, rather than limiting the number of reviewers per file to 1 as in getPreferredReviewers. This is necessary for CheckInternal to check that code reviewers with expertise over specific docs paths are counted as valid docs reviewers.
• Changes getDocsReviewers, CheckInternal, and checkInternalDocsReviews to take a slice of PullRequestFiles so they can apply the getPreferredReviewers logic.

[ptgott]

For context, this github-terraform PR adds preferred reviewers for docs, but it looks like we need to edit the workflow bot for this change to take effect: https://github.com/gravitational/github-terraform/pull/694/files

[zmb3]

Feels like this can result in the same reviewer being added multiple times - is this okay?

[ptgott]

Oversight on my part! I've changed this to deduplicate preferredReviewers. The triple for loop isn't the greatest (it also assumes the number of file paths and reviewers per path remains small), so I've explained the logic in a comment.